• Home
• Articles Archive
• Features List

• Media Pack
• Contact Us
• Subscribe

X-RAY VISIONS

From STORAGE Magazine Vol 7, Issue 2 - March/April 2007

One of the most ambitious medical applications of grid computing is the EU-funded MammoGrid project. SATA technology is helping to deliver the storage performance that is driving its success.

Grid computing is a much-hyped technology, allowing many low-cost computers to work together on complex problems - for example, the analysis of X-rays for medical diagnostics in the European MammoGrid project.

A major challenge in such a system is selecting a solution that will cost effectively store high volumes of image data and retain the performance needed to supply these images to the grid at the data rates demanded. The good news is that AMCC 3ware SATA RAID controllers can be used to create high-performance local image stores to hold the valuable data for this ground-breaking project.

The EU-funded MammoGrid project is undoubtedly one of the most ambitious medical applications of grid computing. The aim is to develop new applications and to investigate the potential of grid technology to improve healthcare. As part of the project, a database of mammograms will be shared across national borders in the EU, creating a rich data resource that will help scientists and researchers throughout Europe.

The vast amount of medical image data in the MammoGrid database allows scientists to perform epidemiological studies, advanced image processing and radiographic education. Ultimately, the goal is to begin 'tele-diagnosis' over communities of medical 'virtual organisations'.

The system manages massively distributed files of mammograms, handles the distributed execution of mammogram analysis software, aids the development of grid-aware algorithms and enables sharing of resources between multiple collaborating medical centres. The MammoGrid uses a software and hardware infrastructure that is specially developed to ensure the integrity and security of the data. The framework is based on AliEn, a grid framework developed by the ALICE Collaboration. This technology provides a virtual file catalogue that allows transparent access to distributed data sets and provides top to bottom implementation of a lightweight grid applicable to cases where the handling of a large number of files is required.

Zybert Computing supplies nodes - also called GridBoxes - to the MammoGrid project. Specialising in high-capacity storage solutions, the company has provided storage for applications that include nuclear physics research, a similar application where distributed computing power is often used to work on enormous data sets.
Although mammograms are monochrome images, the high resolution required means that each file still takes 20MB of storage space. Zybert quickly realised that, although the system capacity and performance would be key factors, one of the most important criteria would be to maximise the storage capacity per square meter of floor space. At many locations where the MammoGrid nodes would be installed, floor space in server rooms is at a premium, so reducing the physical size of the nodes would also substantially reduce the overall project cost.

Like many grid implementations, it was decided that x86 Linux servers would be used for the nodes of the grid, optimising both cost and performance. The nodes also needed to be modular, allowing faulty modules to be replaced easily, ensuring that the on-going costs of the project were minimised.

Each node in the MammoGrid system needed to perform three key functions: storage of images, transmission of images to other nodes and local processing of the mammogram data. The choice of storage technology was critical to ensuring that the capacity, cost and performance were optimised. The system required that each node could transfer up to four of the 20MB X-rays per second (80MB/s) and there was also a requirement to provide a transfer rate to the special high-resolution display screens of 128MB/s.

Conventional wisdom might suggest that SCSI would provide the best trade-off between cost and performance for such an application. However, when the engineers at Zybert began developing the systems, they quickly realised that SATA might provide a better solution, because the data was highly sequential in nature. In many imaging applications - from digital film editing to recording security camera footage - SATA's inherent high transfer rates for sequential data makes it an ideal technology. It is only in applications with high levels of random access applications - such as enterprise databases - that the higher spin-speeds and the larger number of spindles used, due to the lower capacities, make SCSI a better option. Additionally, the low cost per gigabyte of the SATA drives made the final systems very cost effective.

The engineers at Zybert recognised that a hardware RAID controller was needed to ensure that the load on the host processors was minimised. 3ware RAID controllers from AMCC were chosen for the system: in addition to offering the highest performance in the industry, open-source Linux drivers were available for the 3ware controllers, unlike most of the other SATA RAID controllers. The systems were configured with two 80GB drives in a hardware RAID 1 array for the operating system and six 250GB disks for data using both RAID 1 and RAID 0. Four disk bays are available for further expansion. Also, each GridBox has dual XEON 2.8GHz CPUs, 533MHz FSB, 2GB of ECC DDR memory and triple hot-swap 500VA PSUs in a 3U rack-mount case.

To date, the MammoGrid endeavour has been hugely successful as a pilot project, with four sites in Italy, the UK and Switzerland. The implementation phase is expected to involve around 1,000 servers to be installed in European hospitals. As well as continuing to supply the GridBox for medical research, Zybert is also looking at other applications, and is a founder member of the GridBox Consortium that aims to promote Grid computing and Zybert's GridBox to applications such as particle physics and radio astronomy. SATA RAID is the technology that will meet the demands for cost-effective, high-capacity storage for these exciting new applications. ST